applicableInState actions about burlap HOT 2 CLOSED

Yes, isApplicableInState is a means for defining action preconditions and one of the main reasons to do that is so you can provide background knowledge that prunes away impossible actions. Another reason (in the single agent case more so) might be that you have a subsystem that the action calls (like a robotic controller) and you don't want the controller to be initiated in certain conditions.

Adding a definable InvalidActionHandler is an interesting idea, but I'm hesitant to add it. Generally speaking if you have a system that forces execution of an inapplicable action, the system is doing something wrong and it should be alerted. Noop replacements might work in some cases, but other problems could crop up later. For example, if a reward function is defined in such a way that it does not handle the value for inapplicable actions of the domain you could get weird results; in one case the reward function might just break entirely (causing an uninformative runtime exception) or, worse, it might return a nonsensical reward value that that motivates strange unexpected behavior in the agent.

That said, your point about things like visual/terminal explorers not behaving gracefully upon user error is a good point. However, this seems more like a problem that can be addressed in the code of the explorers. That is, the code should simply check whether the user input action is applicable and do nothing, along with a message printed somewhere, if it's not.

Are there other reasons beyond the explorers that you think an InvalidActionHandler would be useful?

from burlap.

FYI, in the latest pull to master, I also made it so that the explorers now do more gracefully handled attempted execution of inapplicable actions.

from burlap.